LessWrong 2.0 Reader

So I keep seeing takes about how to tell if LLMs are "really exhibiting goal-directed behavior" like a human or whether they are instead "just predicting the next token". And, to me at least, this feels like a confused sort of question that misunderstands what humans are doing when they exhibit goal-directed behavior.

Concrete example. Let's say we notice that Jim has just pushed the turn signal lever on the side of his steering wheel. Why did Jim do this?

The goal-directed-behavior story is as follows:

• Jim pushed the turn signal lever because he wanted to alert surrounding drivers that he was moving right by one lane
• He wanted to alert drivers that he was moving one lane right because he wanted to move his car one lane to the right.
• He wanted to move his car one lane to the right in order to accomplish the goal of taking the next freeway offramp
• He wanted to take the next freeway offramp because that was part of the most efficient route from his home to his workplace
• He wanted to go to his workplace because his workplace pays him money
• Jim wants money because money can be exchanged for goods and services
• Jim wants goods and services because they get him things he terminally values like mates and food

But there's an alternative story:

• Jim has had success doing similar things to the successful people in his life
• Jim has learned that "have a job" is the sort of thing that successful people in his life do
• "Showing up for work" is the expected behavior when one has a job
• When one is in a context where the current task is "show up for work", the expected behavior is "drive from home to work by following the directions from Waze"
• When in the context of "I am at my work's exit and I am going to work now", the expected action is "get into the exit lane".
• When in the context of "I'm trying to move one lane to the right", the action "turn on the turn signal" is expected.

I think this latter framework captures some parts of human behavior that the goal-directed-behavior framework misses out on. For example, let's say the following happens

1. Jim is going to see his good friend Bob on a Saturday morning
2. Jim gets on the freeway - the same freeway, in fact, that he takes to work every weekday morning
3. Jim gets into the exit lane for his work, when Bob's house is still many exits away
4. Jim finds himself pulling onto the street his workplace is on

This sequence of actions is pretty nonsensical from a goal-directed-behavior perspective, but is perfectly sensible if Jim's behavior is mostly driven by contextual heuristics.

Note that I'm not saying "humans never exhibit goal-directed behavior".

Instead, I'm saying that "take a goal, and come up with a plan to achieve that goal, and execute that plan" is, itself, just one of the many contextually-activated behaviors humans exhibit.

I see no particular reason that an LLM couldn't learn to figure out when it's in a context like "the current context appears to be in the execute-the-next-step-of-the-plan stage of such-and-such goal-directed-behavior task", and produce the appropriate output token for that context.

Pasteur had (also highly "counterfactual") help I think! He worked in this maternity ward where the women & babies kept dying.  The hospital had opened up some investigations over the years as to the cause of death but kept closing them with garbage explanations. He went somewhere else for a while and when he got back he noticed that the death numbers were down in his absence. Then he noticed his hands smelled like death after one of his routine autopsies and he was about to go plunge them in some poor mother! He had washed them but just with regular soap. If he put some bleach in the washwater then his hands didn't stink. He connected the dots. He had killed hundreds of mothers & babies but wrote a book about it anyway and thereby popularized disinfection (and strongly suggested the root cause of disease).

Probably the main reason that germ theory took so long to work out is that the people with the right evidence were too guilty and ashamed to share it.

I wonder how much testosterone during puberty lowers IQ. Most of my high school math/CS friends seemed low-T and 3/4 of them transitioned since high school. They still seem smart as shit. The higher-T among us seem significantly brain damaged since high school (myself included). I wonder what the mechanism would be here...

Like 50% of my math/cs Twitter is trans women and another 40% is scrawny nerds and only like 9% big bald men.

I have a tremendously large skull (like XXL hats) - maybe that's why I can still do some basic math after the testosterone brain poison during puberty? My voice is kind of high pitched for my body — related?? My big strong brother got the most brain damaged and my thin brother kept most of what he had.

Now I'm looking at tech billionaires. Mostly lo-T looking men. Elon Musk & Jeff Bezos were big & bald but seem to have pretty big skulls  to compensate ¿

I guess this topic/theory is detested by cis women, trans women,  lo-T men, and hi-T men all alike because it has something bad to say about all of them.

 But here's a recipe for success according to the theory:

• be born with giant head (please don't kill your mother, maybe suggest she get a C section)
• delay your puberty until you've learned enough to get by, maybe age 22 or so
• start slamming testosterone and amphetamines to get your workaholicism, betterThanEveryone complex, and drive for power
•  go to Turkey for a hair transplant
•  profit

(I appreciate object-level engagement in general, but this seems combatively worded.)

The rest of this reply responds to arguments.

Why should the Earth superintelligence care about you, but not about the other 10^10^30 other causally independent ASIs that are latent in the hypothesis space, each capable of running enormous numbers of copies of the Earth ASI in various scenarios?

• The example talks of a single ASI as a toy scenario to introduce the central idea.
  • The reader can extrapolate that one ASI's actions won't be relevant if other ASIs create a greater number of copies.
  • This is a simple extrapolation, but would be difficult for me to word into the post from the start.
• It sounds like you think it would be infeasible/take too much compute for an ASI to estimate the distribution of entities simulating it, given the vast amount of possible entities. I have some probability on that being the case, but most probability on there being reasons for the estimation to be feasible:
  • e.g if there's some set of common alignment failure modes that occur across civilizations, which tend to produce clusters of ASIs with similar values, and it ends up being the case that these clusters make up the majority of ASIs.
  • or if there's a schelling-point for what value function to give the simulated copies, that many ASIs with different values would use precisely to make the estimation easy. E.g., a value function which results in an ASI being created on Earth which then gathers more compute than would be had locally, uses it to estimate the distribution of ASIs which engaged in this, and then maximizes the mix of their values.
    • (I feel confident (>90%) that there's enough compute in a single reachable-universe-range to do the estimation, for reasons that are less well formed, but one generating intuition is that I can already reason a little bit about the distribution of superintelligences, as I have here, with the comparatively tiny amount of compute that is me)

On your second paragraph: See the last dotpoint in the original post, which describes a system ~matching what you've asserted as necessary, and in general see the emphasis that this attack would not work against all systems. I'm uncertain about which of the two classes (vulnerable and not vulnerable) are more likely to arise. It could definitely be the case that the vulnerable class is rare or almost never arises in practice.

But I don't think it's as simple as you've framed it, where the described scenario is impossible simply because a value function has been hardcoded in. The point was largely to show that what appears to be a system which will only maximize the function you hardcoded into it could actually do something else in a particular case -- even though the function has indeed been manually entered by you.

Seems it is easier / more streamlined / more googlable now for a teenage male to get testosterone blockers than testosterone. Latter is very frowned upon — I guess because it is cheating in sports. Try googling eg "get testosterone prescription high school reddit -trans -ftm". The results are exclusively people shaming the cheaters. Whereas of course googling "get testosterone blockers high school reddit" gives tons of love & support & practical advice.

Females however retain easy access to hormones via birth control.

But if your goal is to achieve high counterfactual impact in your own research, then you should probably draw inspiration from the opposite: "singular" discoveries, i.e. discoveries which nobody else was anywhere close to figuring out.

This idea reminds me of the concepts in this post: Focus on the places where you feel shocked everyone's dropping the ball [LW · GW].

The LessWrong Review [? · GW] runs every year to select the posts that have most stood the test of time. This post is not yet eligible for review, but will be at the end of 2025. The top fifty or so posts are featured prominently on the site throughout the year. Will this post make the top fifty?

Often I see people dismiss the things the Epicureans got right with an appeal to their lack of the scientific method, which has always seemed a bit backwards to me.

The most important thing, I think, is not even hitting the nail on the head, but knowing (i.e. really acknowledging) that a nail can be hit in multiple places. If you know that, the rest is just a matter of testing.

I think what you proved essentially boils down to the fact that a consistent guessing oracle can be used to compute a completion of any consistent recursively axiomatizable theory. (In fact, it turns out that a consistent guessing oracle can be used to compute a model (in the sense of functions and relations on a set) of any consistent recursively axiomatizable theory; this follows from what you showed and the fact that an oracle for a complete theory can be used to compute a model of that theory.)

I disagree with

Philosophically, what I take from this is that, even if statements in a first-order theory such as Peano arithmetic appear to refer to high levels of the Arithmetic hierarchy, as far as proof theory is concerned, they may as well be referring to a fixed low level of hypercomputation, namely a consistent guessing oracle.

The translation from T to U is computable. The consistent guessing oracle only came in to find a completion of U, but it could also find a completion of T (in fact, a completion of U can be computably translated to a completion of T), so the consistent guessing oracle doesn't really have anything to do with the relationship between T and U.

I prefer to keep plans private but I'm making big progress on meditation and mental re-wiring. Am working on a way to publicly demonstrate. Public plans just stress me out. I recently set two pretty ambitious goals. I figured I could use psychedelics to turbo-charge progress. The meditation one is coming along FAST.

The other goal is honestly blocked a bit on being super out of shape. Multiple rounds of covid really destroyed my cardio and energy levels. Need to rebuild those before a big push on goal 2.